The current frontend implementation closes the connection once `workflow_paused` SSE event is received and establish a new connection to subscribe new events. The implementation of `StreamsBroadcastChannel` sets initial `_last_id` to `0-0`, consumes streams from start and send `workflow_paused` event created before pauses to frontend, causing excessive connections being established.
This PR fixes the issue by setting initial id to `$`, which means only new messages are received by the subscription.
The `/api/system-features` is required for the web app initialization.
Authentication would create circular dependency (can't authenticate without web app loading).
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
The `/console/api/system-features` is required for the dashboard initialization. Authentication would create circular dependency (can't login without dashboard loading).
ref: CVE-2025-63387
Related: #31368
The original fix seems correct on its own. However, for chatflows with multiple answer nodes, the `message_replace` command only preserves the output of the last executed answer node.
- Ensure `EventManager._notify_layers` logs exceptions instead of silently swallowing them
so GraphEngine layer failures surface for debugging
- Introduce unit tests to assert the logger captures the runtime error when collecting events
- Enable the `S110` lint rule to catch `try-except-pass` patterns
- Add proper error logging for existing `try-except-pass` blocks.
This commit:
1. Convert `pause_reason` to `pause_reasons` in `GraphExecution` and relevant classes. Change the field from a scalar value to a list that can contain multiple `PauseReason` objects, ensuring all pause events are properly captured.
2. Introduce a new `WorkflowPauseReason` model to record reasons associated with a specific `WorkflowPause`.
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: -LAN- <laipz8200@outlook.com>
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Certain metadata (including but not limited to `InvokeFrom`, `call_depth`, and `streaming`) is required when resuming a paused workflow. However, these fields are not part of `GraphRuntimeState` and were not saved in the previous
implementation of `PauseStatePersistenceLayer`.
This commit addresses this limitation by introducing a `WorkflowResumptionContext` model that wraps both the `*GenerateEntity` and `GraphRuntimeState`. This approach provides:
- A structured container for all necessary resumption data
- Better separation of concerns between execution state and persistence
- Enhanced extensibility for future metadata additions
- Clearer naming that distinguishes from `GraphRuntimeState`
The `WorkflowResumptionContext` model makes extending the pause state easier while maintaining backward compatibility and proper version management for the entire execution state ecosystem.
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
This PR introduces a `BroadcastChannel` abstraction with broadcasting and at-most once delivery semantics, serving as the communication component between celery worker and API server.
It also includes a reference implementation backed by Redis PubSub.
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>
Disable SSE events truncation for service api invocations to ensure backward compatibility.
Co-authored-by: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com>
The `ChatMessageApi` (`POST /console/api/apps/{app_id}/chat-messages`) and
`ModelConfigResource` (`POST /console/api/apps/{app_id}/model-config`)
endpoints do not properly validate user permissions, allowing users without `editor`
permission to access restricted functionality.
This PR addresses this issue by adding proper permission check.
This PR fixes Alembic offline mode (`--sql` flag) by ensuring data migration functions only execute in online mode. When running in offline mode, these functions now skip data operations and output informational comments to the generated SQL.
The `Account._current_tenant` object is loaded by a database session (typically `db.session`) whose lifetime
is not aligned with the Account model instance. This misalignment causes a `DetachedInstanceError` to be raised
when accessing attributes of `Account._current_tenant` after the original session has been closed.
To resolve this issue, we now reload the tenant object with `expire_on_commit=False`, ensuring the tenant remains
accessible even after the session is closed.
Fix regression introduced in PR #22580 where admin users encountered
"Message not exists" errors when creating feedback on messages created
by other users.
The issue was caused by `MessageService.create_feedback()` incorrectly
filtering messages by the current user's ID, preventing admins from
accessing messages created by end users.
Reverts: #22580
This commit introduces a background task that automatically deletes `WorkflowDraftVariable` records when
their associated workflow apps are deleted.
Additionally, it adds a new cleanup script
`cleanup-orphaned-draft-variables` to remove existing orphaned draft variables from the database.
Fix flaky test
`TestWorkflowDraftVariableService.test_list_variables_without_values_success`
caused by low entropy in test data generation that led to
duplicate values violating unique constraints.
Also improve data generation in other tests within
`TestWorkflowDraftVariableService` to reduce the likelihood of
generating duplicate test data.
This PR introduces UUIDv7 implementations in both Python and SQL to establish the foundation for migrating from UUIDv4 to UUIDv7 as proposed in #19754.
ID generation algorithm of existing models are not changed, and new models should use UUIDv7 for ID generation.
Close#19754.
refactor(api): Separate SegmentType for Integer/Float to Enable Pydantic Serialization (#22025)
This PR addresses serialization issues in the VariablePool model by separating the `value_type` tags for `IntegerSegment`/`FloatSegment` and `IntegerVariable`/`FloatVariable`. Previously, both Integer and Float types shared the same `SegmentType.NUMBER` tag, causing conflicts during serialization.
Key changes:
- Introduce distinct `value_type` tags for Integer and Float segments/variables
- Add `VariableUnion` and `SegmentUnion` types for proper type discrimination
- Leverage Pydantic's discriminated union feature for seamless serialization/deserialization
- Enable accurate serialization of data structures containing these types
Closes#22024.
This PR fix the issue that `ObjectSegment` are not recursively added to the draft variable pool while loading draft variables from database. It also fixes an issue about loading variables with more than two elements in the its selector.
Enhances #19735.
Closes#21477.
This PR addresses issue #21441 by implementing explicit `version` method definitions for all `BaseNode` subclasses to improve code maintainability.
### Changes
Added explicit `version` method definitions for all `BaseNode` subclasses:
- `QuestionClassifierNode`
- `KnowledgeRetrievalNode`
- `AgentNode`
Added comprehensive test suite to validate:
1. All subclasses of `BaseNode` have explicitly defined `version` method
2. All subclasses have required `_node_type` property
3. The `(node_type, node_version)` combination is unique across all subclasses
This pull request introduces a feature aimed at improving the debugging experience during workflow editing. With the addition of variable persistence, the system will automatically retain the output variables from previously executed nodes. These persisted variables can then be reused when debugging subsequent nodes, eliminating the need for repetitive manual input.
By streamlining this aspect of the workflow, the feature minimizes user errors and significantly reduces debugging effort, offering a smoother and more efficient experience.
Key highlights of this change:
- Automatic persistence of output variables for executed nodes.
- Reuse of persisted variables to simplify input steps for nodes requiring them (e.g., `code`, `template`, `variable_assigner`).
- Enhanced debugging experience with reduced friction.
Closes#19735.
- Extract methods used by `ParameterExtractorNode` from `LLMNode` into a separate file.
- Convert `ParameterExtractorNode` into a subclass of `BaseNode`.
- Refactor code referencing the extracted methods to ensure functionality and clarity.
- Fixes the issue that `ParameterExtractorNode` returns error when executed.
- Fix relevant test cases.
Closes#20840.
The `QuestionClassifierNode` class extends `LLMNode`, meaning that, per the Liskov Substitution Principle, `QuestionClassifierNodeData` **SHOULD** be compatible in contexts where `LLMNodeData` is expected.
However, the absence of the `structured_output_enabled` attribute violates this principle, causing `QuestionClassifierNode` to fail during execution.
This commit implements a quick and temporary workaround. A proper resolution would involve refactoring to decouple `QuestionClassifierNode` from `LLMNode` to address the underlying design issue.
Fixes#20725.
- Add `node_execution_id` column to `WorkflowDraftVariable`, allowing efficient implementation of
the "Reset to last run value" feature.
- Add additional index for `WorkflowNodeExecutionModel` to improve the performance of last run lookup.
Closes#20745.
Currently, `WorkflowNodeExecution.execution_metadata_dict` returns `None` when metadata is absent in the database. This requires all callers to perform `None` checks when processing metadata, leading to more complex caller-side logic.
This pull request updates the `execution_metadata_dict` method to return an empty dictionary instead of `None` when metadata is absent. This change would simplify the caller logic, as it removes the need for explicit `None` checks and provides a more consistent data structure to work with.
QuantumGhost
